Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Autonomous Question Formation for Large Language Model-Driven AI Systems

Hong Su

TL;DR

The paper tackles the problem that autonomous LLM-driven systems in dynamic environments struggle to determine which problems to pursue. It introduces a human-simulation framework where autonomous question formation precedes task execution, enabled by progressively enriched prompting scopes that incorporate internal state, environmental perception, and inter-agent awareness, and a learnable question-formation policy derived from experience. The approach is validated in a controlled multi-agent simulation, showing that environment-aware prompting reduces short-term no-eat events over baseline, and adding inter-agent awareness yields a further substantial reduction in cumulative no-eat events over a 20-day horizon, with statistical significance. Collectively, the work identifies autonomous question formation as a fundamental capability, demonstrates a practical framework for integrating internal, external, and social information, and demonstrates meaningful improvements in sustainability and decision quality for open, multi-agent systems.

Abstract

Large language model (LLM)-driven AI systems are increasingly important for autonomous decision-making in dynamic and open environments. However, most existing systems rely on predefined tasks and fixed prompts, limiting their ability to autonomously identify what problems should be solved when environmental conditions change. In this paper, we propose a human-simulation-based framework that enables AI systems to autonomously form questions and set tasks by reasoning over their internal states, environmental observations, and interactions with other AI systems. The proposed method treats question formation as a first-class decision process preceding task selection and execution, and integrates internal-driven, environment-aware, and inter-agent-aware prompting scopes to progressively expand cognitive coverage. In addition, the framework supports learning the question-formation process from experience, allowing the system to improve its adaptability and decision quality over time. xperimental results in a multi-agent simulation environment show that environment-aware prompting significantly reduces no-eat events compared with the internal-driven baseline, and inter-agent-aware prompting further reduces cumulative no-eat events by more than 60% over a 20-day simulation, with statistically significant improvements (p < 0.05).

Autonomous Question Formation for Large Language Model-Driven AI Systems

TL;DR

The paper tackles the problem that autonomous LLM-driven systems in dynamic environments struggle to determine which problems to pursue. It introduces a human-simulation framework where autonomous question formation precedes task execution, enabled by progressively enriched prompting scopes that incorporate internal state, environmental perception, and inter-agent awareness, and a learnable question-formation policy derived from experience. The approach is validated in a controlled multi-agent simulation, showing that environment-aware prompting reduces short-term no-eat events over baseline, and adding inter-agent awareness yields a further substantial reduction in cumulative no-eat events over a 20-day horizon, with statistical significance. Collectively, the work identifies autonomous question formation as a fundamental capability, demonstrates a practical framework for integrating internal, external, and social information, and demonstrates meaningful improvements in sustainability and decision quality for open, multi-agent systems.

Abstract

Large language model (LLM)-driven AI systems are increasingly important for autonomous decision-making in dynamic and open environments. However, most existing systems rely on predefined tasks and fixed prompts, limiting their ability to autonomously identify what problems should be solved when environmental conditions change. In this paper, we propose a human-simulation-based framework that enables AI systems to autonomously form questions and set tasks by reasoning over their internal states, environmental observations, and interactions with other AI systems. The proposed method treats question formation as a first-class decision process preceding task selection and execution, and integrates internal-driven, environment-aware, and inter-agent-aware prompting scopes to progressively expand cognitive coverage. In addition, the framework supports learning the question-formation process from experience, allowing the system to improve its adaptability and decision quality over time. xperimental results in a multi-agent simulation environment show that environment-aware prompting significantly reduces no-eat events compared with the internal-driven baseline, and inter-agent-aware prompting further reduces cumulative no-eat events by more than 60% over a 20-day simulation, with statistically significant improvements (p < 0.05).
Paper Structure (46 sections, 45 equations, 4 figures)

This paper contains 46 sections, 45 equations, 4 figures.

Table of Contents

  1. Introduction
  2. Related Work
  3. LLM-Driven Agents and Prompt-Based Decision Making
  4. Autonomous Decision-Making in Dynamic and Open Environments
  5. Multi-Agent Interaction and Socially Aware Systems
  6. Summary and Distinction
  7. To Find Questions and Set Up Tasks Automatically in a Human-Simulation Way
  8. Internal Questions and Tasks
  9. Goal of the Autonomous AI System
  10. Sensor Data of the AI System Itself
  11. Self-Discovered Importance and Indirect Internal Questions
  12. Outer-World Questions and Tasks
  13. Difference-Based Screening
  14. Strong or Abnormal Measurements in the Outer World
  15. Full-Scope Screening When Computation Is Sufficient
  16. ...and 31 more sections

Figures (4)

  • Figure 1: Daily number of no-eat events for the three methods over the 20-day simulation period. Environment-aware and inter-agent-aware strategies substantially reduce short-term resource shortages compared with the internal-driven baseline.
  • Figure 2: Cumulative number of no-eat events over time. Method 3 consistently achieves the lowest cumulative no-eat count, indicating superior long-term sustainability through environmentally and inter-agent-informed decision-making.
  • Figure 3: Daily number of greeting ("hello") actions. Method 3 exhibits active and sustained inter-agent interaction, which prevents social-condition violations and indirectly supports stable environmental maintenance.
  • Figure 4: Daily number of watering actions. Method 2 and Method 3 significantly increase watering frequency compared with the baseline, accelerating vegetable regeneration. Method 3 achieves a more stable watering pattern due to inter-agent constraints.